Compute Kit contents
What is in the Kit and which parts you need for your task, with pointers to the indexes that list exact paths.
Start with the guide for your task. It shows which Kit parts you need and where to read next. Use the complete file index when you need a particular implementation.
During public review, read the linked guides. Artifact paths refer to a separately supplied, authorized Kit. The review does not install software, transfer task input, or submit paid work; those actions follow their own access and approval steps.
The Kit is the harness around your existing agent: it supports permitted preparation, approved connection and result handling. Your agent keeps context, planning and final decisions. Participating Macs process admitted work; an authorized monthly report is a separate reporting API or file path.
Choose by task
| What you need | Start here | Read next |
|---|---|---|
| Prepare permitted data | Preparation | Review the Kit |
| Review code | Code & Tests | Use the agent guide |
| Extract document information | Documents & Search | Use the agent guide |
| Translate content | Translation | Use the agent guide |
| Connect your agent | Use Compute with your AI | Setup instructions in your approved Kit |
| Check or save a result | Results | Read and verify result files |
For the five information families and eight branded workflow guides, use Choose work. That taxonomy helps choose a path; the selected service still decides its accepted format, operation, limits and output.
Read the supplied indexes selectively
The supplied Kit includes 559 files. Use docs/FILE_INDEX.md to browse the complete catalog or catalog/kit-index.json to select exact files by task. Paths are relative to the extracted Kit.
Choose a language or host
The Kit exports TypeScript, Python, Go, Swift and .NET SDK families. TypeScript includes the complete four-module path. Python includes selected record, format, result-bundle and impact helpers. Go, Swift and .NET are chiefly wire clients. A package name does not imply installation from a public package registry. Use the SDK parity guide and the interface reference for the selected contract.
For agent connections, the five remote work tools are lithi.quote, lithi.run, lithi.batch,
lithi.status and lithi.result. They are distinct from SDK methods, local helpers and skills.
Claude Code, Codex and Cursor can use portable skills and MCP where the selected host supports
those surfaces. Host support does not prove local file access, key custody or a result destination;
the host guide owns current host details.
Keep preparation and approval visible
Use data preparation to decide what may leave your machine.
The local decision is blocked, review_required or eligible. A blocked run withholds the
original input and does not silently send a fallback. Pseudonymization can reduce identifying
information, but the protected reverse map stays in the customer environment and outside the
agent's reasoning context. If policy forbids participant plaintext from leaving the permitted
environment, the selected work is a no-fit.
The Kit supplies guidance and local composition. It does not grant access or permission. A quote prices a declared scope; transfer permission and execution approval remain separate. For the full journey from task fit through result checks and monthly report interpretation, continue with the agent guide.